Locality Preserving The Canonical Correlation Analysis And Its Application To Face Recognition

As a new type of algorithm, locality preserving canonical correlation analysis (LPCCA) can solve a large number of non-linear problems. It solves nonlinear problems through local linearization method; it not only maintains sample set of local structural information, but also maximizes the relevant information between two sets of samples. In this article, LPCCA is applied to face recognition; we advances two improved algorithms through the analysis of its inadequacy in classification. Our work can be divided into four parts as below:(1) LPCCA is applied in face recognition. First, in order to obtain the necessary sample sets which LPCCA and new algorithms need, the features of the source images are extracted through the principal component analysis and two-dimensional discrete wavelet transform. Then we discuss the effectiveness of LPCCA through experiment of face recognition, analyze its deficiencies and give the direction of improvement.(2) Improved locality preserving canonical correlation analysis. Based on LPCCA, and through introducing the categories of information, we advance improved locality preserving canonical correlation analysis (ILPCCA). It not only maintains local structural information between samples of within-class, but also maximizes the relevant information between two sets of samples. Face recognition experiment proves that it has higher recognition performance.(3) Improved local preserving canonical correlation analysis. In this paper, based on discriminative canonical correlation analysis (DCCA) and locality discriminative canonical correlation analysis (LDCCA), through introducing the categories of information fully, we advance improved local preserving canonical correlation analysis. It not only maintains local structural information between samples of one category, but also can realize the maximization of local within-class correlations and the minimization of local between-class correlations. Experiments show that, ILDCCA has higher recognition rate and stability.(4) Determination of the parameter in neighbor samples. The same as locality preserving projection and LPCCA, ILPCCA and ILDCCA have the problem of determining the parameter in neighbor samples. In this paple, we give an experience value by experient.